Evaluation of AutomatedPedestrian Detection atSignalized IntersectionsREPORT NO. FHWA-RD-00-097 August 2001

U.S. Department of TransportationFederal Highway AdministrationResearch, Development, and TechnologyTurner-Fairbank Highway Research Center6300 Georgetown PikeMcLean, VA 22101-2296

FOREWORD The FHWA’s Pedestrian and Bicycle Safety Research Program’s overall goal is to increase pedestrian and bicycle safety and mobility. From better crosswalks, sidewalks and pedestrian technologies to growing educational and safety programs, the FHWA’s Pedestrian and Bicycle Safety Research Program strives to pave the way for a more walkable future. At many signalized intersections, pedestrian detection is accomplished by the pedestrians pushing buttons to activate the Walk phase. Pedestrians often do not activate the Walk signal because they believe the button already has been pressed, do not think it is working, or cannot access the button. Automated pedestrian detection systems can detect the presence of pedestrians and call the Walk signal without any require action by the pedestrian. This study evaluated the safety effects of microwave and infrared detectors used in conjunction with standard pedestrian push buttons. This study was part of a larger Federal Highway Administration research study investigating the effectiveness of innovative engineering treatments on pedestrian safety. It is hoped that readers also will review the reports documenting the results of the related pedestrian safety studies. The results of this research will be useful to transportation engineers, planners, and safety professionals who are involved in improving pedestrian safety and mobility.

Michael F. Trentacoste

Director, Office of Safety Research and Development

NOTICE This document is disseminated under the sponsorship of the Department of Transportation in the interest of information exchange. The United States Government assumes no liability for its content of use thereof. This report does not constitute a standard, specification or regulation. The United States Government does not endorse products or manufactures. Trade and manufactures’ names appear in this report only because they are considered essential to the object of the document.

Technical Report Documentation Page 1. Report No.

FHWA-RD-00-0972. Government Accession No. 3. Recipient's Catalog No.

4. Title and Subtitle

EVALUATION OF AUTOMATED PEDESTRIAN DETECTION AT SIGNALIZED INTERSECTIONS

5. Report Date

August 20016. Performing Organization Code

7. Author(s)

Ronald Hughes, Herman Huang, Charles Zegeer, and Michael Cynecki

8. Performing Organization Report No.

9. Performing Organization Name and Address

Highway Safety Research Center City of Phoenix University of North Carolina Street Transportation Dept. 730 Airport Rd, Bolin Creek Ctr. 200 West Washington Chapel Hill, NC 27599-3430 Phoenix, AZ 85003-1611

10. Work Unit No. (TRAIS)

11. Contract or Grant No.

DTFH61-92-C-00138

12. Sponsoring Agency Name and Address

Federal Highway Administration Turner-Fairbank Highway Research Center 6300 Georgetown Pike McLean, VA 22101-2296

13. Type of Report and Period Covered

Research Report July 1, 1996 - October 31, 199914. Sponsoring Agency Code

15. Supplementary Notes

Contracting Officer’s Technical Representative (COTR): Carol Tan Esse, HRDS-05 16. Abstract

Automated pedestrian detection systems provide the means to detect the presence of pedestrians as theyapproach the curb prior to crossing the street, and then “call” the Walk signal without any action required on thepart of the pedestrian. The objective of the present study was to evaluate whether automated pedestriandetectors, when used in conjunction with standard pedestrian push buttons, would result in fewer overallpedestrian-vehicle conflicts and fewer inappropriate crossings (i.e., beginning to cross during the Don’t Walksignal). “Before” and “after” video data were collected at intersection locations in Los Angeles, CA (infrared andmicrowave), Phoenix, AZ (microwave), and Rochester, NY (microwave). The results indicated a significantreduction in vehicle-pedestrian conflicts as well as a reduction in the number of pedestrians beginning to crossduring the Don’t Walk signal. The differences between microwave and infrared detectors were not significant. Detailed field testing of the microwave equipment in Phoenix revealed that fine tuning of the detection zone is stillneeded to reduce the number of false calls and missed calls.

17. Key Words

automatic pedestrian detection, microwave, infrared, signals, conflicts

18. Distribution Statement

No restrictions. This document is available to thepublic through the National Technical InformationService, Springfield, VA 22161.

19. Security Classif. (of this report)

Unclassified20. Security Classif. (of this page)

Unclassified21. No. of Pages

2322. Price

Form DOT F 1700.7 (8-72) Reproduction of form and completed page is authorized

ii

TABLE OF CONTENTS

Page

INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1Statement of the Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1Microwave and Infrared Technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1Experience With Automated Pedestrian Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2Objective of the Present Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

METHOD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3Los Angeles, Temple and San Pedro (east side crosswalk) . . . . . . . . . . . . . . . . . . . . . . . . . . 4Rochester, Crittenden and Lattimore (north side crosswalk) . . . . . . . . . . . . . . . . . . . . . . . . . 5Rochester, State and Corinthian (north side crosswalk) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5Phoenix, Central at Earll (south side crosswalk) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

RESULTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8Pedestrians Who Began to Cross During the Steady Don’t Walk . . . . . . . . . . . . . . . . . . . . . 8The Effects of Automated Detection on Pedestrian-Vehicle Conflicts . . . . . . . . . . . . . . . . . . . 9Accuracy of Microwave Sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

DISCUSSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

CONCLUSIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

ACKNOWLEDGMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

APPENDIX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17Evaluation of the SMARTWALK 1400 Microwave Pedestrian Sensor in Phoenix, Arizona . 17

REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

iii

LIST OF FIGURES

Page

Figure 1. An automated pedestrian detection system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

Figure 2. Temple and San Pedro (Los Angeles) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

Figure 3. Crittenden and Lattimore (Rochester) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

Figure 4. State and Corinthian (Rochester) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

Figure 5. Central and Earll (Phoenix) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

Figure 6. Pedestrians who started crossing when the pedestrian signal was steady Don’t Walk and parallel traffic had the green . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

Figure 7. Pedestrians who experienced conflicts with motor vehicles . . . . . . . . . . . . . . . . . . . . . . . 10

Figure 8. Pedestrian-vehicle conflicts as a function of the percentage of pedestrians crossing during the steady Don’t Walk phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

Figure 9. Change in when pedestrians started to cross after automated pedestrian detectors were added . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

Figure 10. Change in pedestrian-motor vehicle conflicts by type after automated pedestrian detectors were added . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

LIST OF TABLES

Page

Table 1. Number of Pedestrians and Hours of Data Collection at Each Study Site . . . . . . . . . . . . . . 4

iv

1

INTRODUCTION

Statement of the Problem

Pedestrian Walk / Don’t Walk signals are special types of traffic control devices intended forcontrolling pedestrian traffic (MUTCD, 1988). The conventional Walk / Don’t Walk messages providepedestrians with reliable information about (a) when it is appropriate to begin crossing the street (steadyWalk signal), (b) when pedestrians should not start crossing (flashing Don’t Walk), and (c) whenpedestrians should not be in the street at all (steady Don’t Walk). To optimize the efficiency of trafficsignals, many are designed to be vehicle-actuated. At actuated traffic signals, pedestrians may have topress a push button in order to receive the Walk signal and to ensure that they will have enough time tocross the street.

A problem with this arrangement is that not all individuals wanting to cross will press the pushbutton. A previous study suggested that only about half of all pedestrians use the push button (Zegeer etal., 1985). There are a number of possible reasons why pedestrians do not use the push button. Theymay not be aware that pressing the button is necessary to obtain the Walk signal, because many signalsdo not have a push button and automatically allocate a Walk interval on every cycle. Even whenpedestrians are aware of the requirement, the delay between the time that the push button is pressed andthe Walk signal appears can be long enough that some pedestrians think that the system ismalfunctioning. Visually impaired pedestrians may not realize that the push button exists or may not beable to find it (Bentzen and Tabor, 1998). Pedestrians with severe mobility impairments may be unableto press a conventional push button. In any case, the result is that pedestrians may attempt to crossagainst the signal.

A number of different automated pedestrian detection technologies have been proposed as ameans of detecting the presence of the pedestrian, so that he / she does not have to press the button. These include the use of infrared, microwave, and video image processing (for example, Zegeer et al.,1994; UK Department of Transport, 1993; Ekman and Draskocsky, 1992; King, 1994).

Microwave and Infrared Technologies

A microwave detector generates a beam of energy at a particular frequency. The basis fordetection is the device’s ability to detect a difference in the frequencies of the outgoing beam and thebeam that is reflected back (the Doppler effect). The beam has to be targeted accurately, especiallywhen the size of the object to be detected (e.g., a pedestrian) is significantly less that of other movingobjects (e.g, passing vehicles).

Infrared technologies are already well established for both vehicle and off-road pedestriandetection. Examples of locations in which infrared is used to detect pedestrian presence include grocerystores, shops, banks, and entrances to other public buildings. The efficiency of infrared detection

2

Figure 1. An automated pedestrian detection system.

methods can be degraded if the object remains still. Infrared devices cannot discriminate the directionof pedestrian movement, nor can they determine the number of objects detected.

Both microwave and infrared detectors work by calling the Walk signal when a person entersthe detection zone on the curb. The size and shape of the detection zone varies depending on the typeof detector used and how it is positioned. A delay can be built in so that persons are detected only ifthey stay within the detection zone for more than a minimum amount of time. Such a delay will help to

prevent false actuations resulting from persons who are merely passing through the detection zone anddo not intend to cross the street. A more detailed discussion of alternative detection technologies isgiven elsewhere (Sherborne, 1992).

Experience With Automated Pedestrian Detection

In the United Kingdom, Puffin (Pedestrian User-Friendly INtelligent) crossings respond topedestrian demand and do not delay traffic unnecessarily when no pedestrians are present (UKDepartment of Transport, 1993). Pedestrian presence is sensed either by use of a pressure- sensitivemat or by an infrared detector mounted above the crossing location. Pressure on the mat is used bothfor initial detection as well as to confirm that the pedestrian has not departed the crossing zone beforethe Walk signal appears. If the pedestrian departs the crossing zone prior to the appearance of theWalk signal, the call will be canceled.

Puffin crossings may also utilize an additional sensor to detect the continued presence ofpedestrians in the crosswalk, thereby allowing the signal phase to be extended for those requiringadditional time to cross. The conversion of a standard signal to a Puffin crossing in Victoria, Australia,reduced by 10 percent the number of pedestrians who started to cross before the pedestrian Walksignal was presented (Catchpole, 1996). Similar results were reported in Växjö, Sweden (Ekman and

3

Draskoczy, 1992). The Swedish results also showed that the number of vehicle-pedestrian conflictsdecreased after the microwave detectors were in place.

The Dutch PUSSYCATS (Pedestrian Urban Safety System and Comfort At Traffic Signals)system consists of a pressure-sensitive mat to detect pedestrians waiting to cross, infrared sensors todetect pedestrians within the crossing, and a near-side pedestrian display (Tan and Zegeer, 1995). Although pedestrians perceived PUSSYCATS to be at least as safe as the old system, manypedestrians reported that they did not understand the function of the mat. As many as half of allpedestrians refused to use the system. Similar applications are being used in the United Kingdom andFrance (Levelt, 1992).

Objective of the Present Study

Although automatic detection systems for pedestrians have been used successfully in Europeand Australia for several years, such applications are in the early stages of testing and use in the UnitedStates. Differences in signal hardware plus cost and other considerations have prevented direct andimmediate application of these technologies in the United States.

The present study evaluated automated pedestrian detection systems installed at sites in LosAngeles, CA (infrared and microwave); Phoenix, AZ (microwave); and Rochester, NY (microwave). At the Los Angeles site, a second set of sensors was used to detect pedestrians who were crossing inthe crosswalk. This additional detector monitored pedestrians in the crosswalk. It extended theclearance interval (steady Don’t Walk before opposing traffic had the green) until pedestrians weresafely on the other side of the street. All other locations were equipped with only one sensor thatdetected pedestrians who arrived in the queuing area at the curb.

The goals of the analysis were to determine if automated pedestrian detectors, when used inconjunction with standard pedestrian push buttons (a) decreased the likelihood of pedestrians beginningto cross during the Don’t Walk signal and (b) decreased vehicle-pedestrian conflicts. This study alsoincluded a detailed analysis of the microwave detection zone and detection accuracy in Phoenix.

METHOD

Data collection consisted of videotaping motorist and pedestrian behavior before and afterautomated detectors were installed in Los Angeles, Rochester, and Phoenix. At each location,pedestrian push buttons already existed and remained operational after the automated detectors wereadded. In Los Angeles, data were collected under three conditions, no automated detector inoperation, infrared detector in operation, and microwave detector in operation.

Data were collected during daylight hours, under dry conditions (no snow or ice was on theground in Rochester). A video camera was set up on the sidewalk, approximately 23 m (75 ft)

4

upstream from the intersection, in Rochester and Phoenix. In Los Angeles, the camera was set up onthe top level of a parking deck. The camera was positioned so that the crosswalk of interest wasvisible, and pedestrians in the crosswalk were walking either toward or away from the camera. Thepedestrian queuing areas, the pedestrian signals and push buttons, and traffic signals for parallel trafficwere also recorded by the camera. Table 1 shows the number of hours of data collection and thenumber of pedestrians that were videotaped at each study site.

Table 1. Number of Pedestrians and Hours of Data Collection at Each Study Site.

STUDY SITE NUMBER OF PEDESTRIANS (HOURS OFDATA COLLECTION IN PARENTHESES)

BEFORE AFTER

Los Angeles: San Pedro and Temple 573 (2 hr) 533 (4 hr - with infrared)590 (1 hr 50 min - withmicrowave)

Rochester: Crittenden and Lattimore 767 (4 hr) 555 (2 hr)

Rochester: State and Corinthian 277 (7 hr 20 min) 208 (3 hr 30 min)

Phoenix: Central and Earll 500 (2 hr) 671 (4 hr)

Los Angeles, Temple and San Pedro (east side crosswalk)

The Temple and San Pedro site is a “T” intersection (Figure 2). Temple Street is an east-weststreet and San Pedro Street comes from the south to form the “T.” There are marked crosswalks andpush buttons across the east and south legs only. A sign directs pedestrians wishing to cross the westleg to the east leg. The signal controlling the south side crossing is on pedestrian recall. Automatedpedestrian detectors were added to the east side crosswalk. On both the northeast and southeastcorners, one detector picked up pedestrians who entered into the detection zone on the sidewalk(waiting to cross the street) and a second detector picked up pedestrians who were still in thecrosswalk.

If a pedestrian was detected as still being in the crosswalk, the crossing time was extended by0.2-s increments to a maximum of 6 additional seconds. The extended crossing time corresponded to awalking speed of 0.9 m/s (3.0 ft/s), compared with the customary 1.2 m/s (4.0 ft/s). During theextended crossing interval, the steady Don’t Walk signal was displayed and the onset of the green signalfor opposing motorists (traveling on Temple) was delayed.

5

Figure 2. Temple and San Pedro (Los Angeles).

Rochester, Crittenden and Lattimore (north side crosswalk)

Lattimore Road is a north-south road, and Crittenden Boulevard is an east-west road (Figure3). The west leg of this intersection is an entrance to a large parking lot. Both the north side and thesouth side crosswalks were controlled by push buttons. Each push button controlled both the north sideand south side crosswalks. The signal controlling the east side crosswalk was on pedestrian recall. Unlike the other three crosswalks, the west side crosswalk was a ladder design, and was not controlledby a pedestrian signal. The north side crosswalk was the treatment site. It connected the parking lot tothe University of Rochester medical complex. The south side crosswalk connects two parking lots.

Rochester, State and Corinthian (north side crosswalk)

This intersection is in downtown Rochester. State Street is a north-south arterial, andCorinthian Street is a very short east-west street that forms a T-intersection to the west (Figure 4). Theeast end of Corinthian is a municipal parking garage. The north side crosswalk is controlled by a pushbutton. There was no pedestrian signal controlling the east side crosswalk. The south leg did not have a

6

Figure 3. Crittenden and Lattimore (Rochester).

Figure 4. State and Corinthian(Rochester).

marked crosswalk. Traffic on State always had the green light unless there was a vehicle on Corinthianor unless a pedestrian pushed the button.

7

Figure 5. Central and Earll (Phoenix).

Phoenix, Central at Earll (south side crosswalk)

Central Avenue is a major arterial with seven lanes and an ADT of more than 38,000 (Figure5). Earll Drive is a main driveway to Park Central Mall (west leg) and aligns with a local street on theeast side of the intersection. The vehicle traffic on Earll is intermittent and the number of pedestrianscrossing Central varies throughout the day. Since traffic volumes on Central are high for most of theday, fixed-time operation (which would provide a pedestrian Walk interval on every cycle, includingtimes when pedestrians are not present) could cause substantial congestion.

8

Figure 6. Pedestrians who started crossing when the pedestrian signal wassteady Don’t Walk and parallel traffic had the green.

RESULTS

The videotapes were watched, and the relevant pedestrian and motorist behaviors wererecorded.

Pedestrians Who Began to Cross During the Steady Don’t Walk

Because automated pedestrian detectors will call the Walk signal for pedestrians who do notpush the button, most pedestrians will have the opportunity to start crossing on the Walk signal. Thus, itwas thought that fewer pedestrians would begin to cross during the steady Don’t Walk after automatedpedestrian detectors were installed. Figure 6 shows the percentage of pedestrians who began to crossduring the Don’t Walk signal under three conditions: (1) push button only, (2) push button andmicrowave detector, and (3) push button and infrared detector.

At the Temple and San Pedro site in Los Angeles, both infrared and microwave detectors were

9

evaluated. Both types of detectors, when used in conjunction with the push button, resulted in asignificant reduction in the percentage of pedestrians beginning to cross during the Don’t Walk signal (P2

= 6.019, df = 1, p = 0.014). Infrared and microwave detectors each produced results that weresignificantly different from the push button only condition (P2 = 6.586, df = 2, p = 0.037). Theeffectiveness of the infrared and microwave detectors did not differ from each other (P2 = 0.867, df = 1,p = 0.352).

In Rochester, at the Crittenden and Lattimore site, the use of the microwave detector togetherwith the push button significantly reduced the number of pedestrians beginning to cross during the Don’tWalk signal (P2 = 43.935, df = 1, p = 0.001). The same results were seen at the Central and Earll sitein Phoenix (P2 = 11.317, df = 1, p = 0.001). A significant change was not, however, achieved with themicrowave detector installed in Rochester at State and Corinthian (P2 = 0.036, df = 1, p = 0.849). Thekey point is that the automated detectors ensure that the Walk display will appear. Failure to use thepush button means that the pedestrian crosses on a steady Don’t Walk display, often without beingassured of sufficient time to get across.

Thus, significant improvements in pedestrian compliance were achieved at three of the four studysites. The magnitude of these effects, while not large, is consistent with that obtained elsewhere(Catchpole, 1996).

With respect to the extended crossing time for pedestrians still in the crosswalk, the data showthe following:

(1) Without automatic pedestrian detection or extension, 47 percent of pedestrians at Temple andSan Pedro finished crossing during the steady Don’t Walk signal while parallel traffic still had agreen. With the extension capability operational, there was a significant increase (from 47 to 53percent) in pedestrians who finished during the steady Don’t Walk while parallel traffic still had agreen (P2 = 5.496, df = 1, p = 0.0l9), even though fewer pedestrians began to cross during thesame steady Don’t Walk signal.

(2) The addition of the extension capability significantly reduced the percentage of pedestrians (from16 percent to 7 percent) who finished crossing during a steady Don’t Walk display after thesignal for oncoming traffic had turned to green (P2 = 42.0l7, df = l, p = 0.00l).

Taken together, these data indicate that the extension capability was serving to increaseprotected time for pedestrians. More pedestrians were able to complete crossing during the (stillprotected) steady Don’t Walk (parallel traffic green). Fewer pedestrians were still in the street duringthe unprotected Don’t Walk (oncoming traffic green).

The Effects of Automated Detection on Pedestrian-Vehicle Conflicts

10

Figure 7. Pedestrians who experienced conflicts with motor vehicles.

In this study, a conflict was defined as any pedestrian-motorist interaction in which either thepedestrian or the motorist stops or slows down so that the other can proceed. For example, all of thefollowing situations were considered to be conflicts:

1. A motorist stops or swerves to avoid hitting a pedestrian.2. A right-turning motorist slows or stops to let a pedestrian cross. 3. A pedestrian crossing against the signal stops halfway across the road to wait for a gap in

opposing traffic.4. Opposing traffic stops and waits for a pedestrian who is crossing against the signal.

Conflict data were collected at Temple and San Pedro in Los Angeles and at the two Rochester sites. The results of the conflict studies are shown in Figure 7.

For the Los Angeles site, the use of automatic pedestrian detectors in conjunction with theconventional push button significantly reduced vehicle-pedestrian conflicts (P2 = 33.533, df = 1, p =0.001). There were no significant differences based on whether the infrared or microwave detector was

11

Conflicts as a Function of Pedestrian Crossing Behavior (logarithmic function)

y = 0.9248x + 11.125R 2 = 0.2031

0

5

10

15

20

25

30

35

0 5 10 15 20Percent of Pedestrians Crossing During Steady DON'T WALK

Per

cen

t of P

edes

tria

ns

Exp

erie

nci

ng

C

on

flic

ts

Figure 8. Pedestrian-vehicle conflicts as a function of the percentage of pedestrians crossing during the steady Don’t Walk phase.

used (P2 = 0.918, df = 1, p = 0.338). Similar effects were obtained with use of microwave detection atCrittenden and Lattimore in Rochester (P2 = 17.653, df = 1, p = 0.001) and at State and Corinthian (P2

= 16.311, df = 1, p = 0.001) in Rochester. Conflict data from the Phoenix site could not be analyzedbecause of visual limitations associated with the site.

Figure 8 shows the relationship between conflicts and the likelihood of pedestrians crossingduring the Don’t Walk phase. A linear relationship accounts for only about 20 percent of the variance(R2 = 0.20). Note that the y-intercept of the function would lead one to predict that about 10 percentof all pedestrians would still encounter some form of conflict even if the percentage of pedestriansbeginning to cross during the Don’t Walk signal was reduced to zero.

Logarithmic and exponential “fits” were found to offer no improvement over the simple linearfunction. While it was possible, for example, to account for almost twice the variance with a fourth-order polynomial, the process that might be responsible for producing such a complex relationship is notat all clear. More precise clarification of these relationships lies beyond the limited data provided byobservations of these three sites. It is sufficient to say that as more pedestrians begin to cross during the

12

Don’t Walk signal, more people will still be in the street when opposing traffic begins to move, andtherefore more conflicts will occur. While conflicts from opposing traffic could be minimized by delayingthe green for opposing traffic, conflicts from turning vehicles would still be present.

The most commonly observed pedestrian action was running, which occurred roughly 10 to 15percent of the time at all sites. This occurred regardless of whether or not the push button wasaugmented with automatic detectors. Running can be a response to seeing the flashing or steady Don’tWalk signal or to seeing the light for parallel traffic change to yellow. Pedestrians may run when theysee traffic approaching from the left or right, or to get out of the path of a turning vehicle.

Accuracy of Microwave Sensors

The discussion above dealt with the effects of the automated detectors on pedestrian behaviorand conflicts. This section discusses detector accuracy in Phoenix; Portland, Oregon; and Los Angeles.

The Phoenix Traffic Engineering Department tested the accuracy of the microwave sensors atCentral and Earll. In this test, a red indicator light unobstrusively located on the signal pole illuminatedwhen the sensor detected a pedestrian waiting to cross. This test was used to determine the number offalse actuations and missed pedestrian calls. False actuations increase the delay to traffic on the mainstreet by creating more stoppages and longer stoppages when there may be no vehicles on the sidestreet nor any pedestrians waiting to cross the main street. Pedestrians who were missed by theautomated detectors could, of course, still call the Walk signal by pressing the button.

The initial testing revealed that the sensors on occasion falsely detected slow-moving, right-turning vehicles next to the pedestrian detection zone. Some of the false actuations could not beexplained by any actions within or near the detection zone. As recommended by the manufacturer, thesystem’s antenna was rotated, and fewer false calls occurred afterwards. Heavy rain during the testperiod resulted in the sensor giving false calls on each cycle.

There were also a few missed calls during the testing period, but these were much fewer innumber than the number of false actuations. The missed calls involved persons who were standingoutside the detection zone. There were no instances of pedestrians standing within the detection zonefor at least 4 s and not being detected.

After various adjustments were made to the sensors, they were activated with the pedestriansignal equipment for full evaluation. In general, while the sensors did not perform correctly in all cases,they had a positive effect upon pedestrian behavior. Experience, however, indicates that further fine-tuning of the equipment could result in additional benefits to pedestrians while minimizing adverse effectson motor vehicle flow.

13

Both microwave and infrared detectors were tested in Portland. The results showed thatmicrowave detectors performed better than infrared detectors in terms of fewer false calls (1percent vs. 4 percent). Microwave detectors were more likely to miss calls (7 percent vs. 1.5 percentfor infrared). The detection zone was larger for the microwave [15 m (49 ft) in length and up to 6 m (20ft) in width] than it was for the infrared [14 m (45 ft) in length and up to 1 m (3 ft) in width] (Beckwithand Hunter-Zaworski, 1998).

The City of Los Angeles tested microwave detectors at two locations, in rainy weather. Thefalse call rate was 3.5 percent. Through traffic in the curb lane and right-turning vehicles triggered thefalse calls. The missed call rate was 1.5 percent. Pedestrians who crossed very close to the sensorpole accounted for the missed calls (Brian Gallagher, Los Angeles Department of Transportation,unpublished data).

More evaluations of this type are recommended to better understand the operational constraintsof new technologies being applied for the first time in the pedestrian environment. A more extensivedescription of the testing in Phoenix is given in the Appendix.

DISCUSSION

The data provide evidence that the use of devices to automatically detect the presence ofpedestrians at signalized crossings can improve the performance at conventional pedestrian push buttonlocations. The data show that improvements can be expected in terms of:

‘ A decrease in the likelihood that pedestrians will begin crossing during the steadyDon’t Walk signal, which decreases the likelihood of encountering opposing traffic.

The addition of automated pedestrian detectors to sites with existing pedestrian push buttonsresulted in an overall 24 percent increase in the number of pedestrians who began to cross during theWalk signal (Figure 9). This was accompanied by an 81 percent decrease overall in the number ofpedestrians who began to cross during the steady Don’t Walk signal. The number of sites upon whichthese results are based is small and, as the data have shown, pedestrian performance can vary widelyacross sites. The reader should therefore be cautious in extrapolating these results to other sites.

Automatic pedestrian detectors seem to derive their effectiveness not from their relationship tothe pedestrian information that is displayed, since the information is identical to that provided when thepush button is pressed. Instead, the improvement is in increasing the likelihood that a pedestrian willreceive the Walk signal. Pedestrian detection (either by the push button being pressed or by automaticdetection) is important in that it ensures that at least a minimum interval will be provided for thepedestrian to cross. A person who crosses without using the push button to “call” the Walk signal hasno way of knowing the duration of the green time for parallel traffic, and therefore is at risk of havinginsufficient time to cross the street.

14

Figure 9. Change in when pedestrians started to cross after automatedpedestrian detectors were added.

‘ A decrease in the number of conflicts experienced by pedestrians while crossing.

The addition of automatic detectors to sites using conventional pedestrian push buttons reducedvehicle-pedestrian conflicts (Figure 10). Conflicts encountered by pedestrians during the first half oftheir crossing were reduced by 89 percent, and conflicts during the second half by 42 percent. Conflictsassociated with right-turning vehicles were reduced by 40 percent. “Other” types of conflicts werereduced by 76 percent.

15

-89

-42

-40

-76

-90

-80

-70

-60

-50

-40

-30

-20

-10

0

PE

RC

EN

T C

HA

NG

E

1st half 2nd half Right-turningvehicle

Other

TYPE OF CONFLICT

Figure 10. Change in pedestrian-motor vehicle conflicts by type after automatedpedestrian detectors were added.

CONCLUSIONS

The present data suggest that automated pedestrian detectors can provide significant operationaland safety benefits when installed in conjunction with conventional pedestrian push buttons at actuatedtraffic signals. The likelihood of fewer inappropriate crossings (during the steady Don’t Walk signal)and fewer vehicle-pedestrian conflicts are expected to be correlated with improved pedestrian safety. Itremains to be shown whether the benefits of automatic detection outweigh its cost. A task that remainsis to correlate the reduction in inappropriate crossings and the reduction in conflicts with a reduction in

16

pedestrian crashes and/or travel delays for both the pedestrian and the motorist. This additionalinformation is needed to develop warrants for the most effective use of these new technologies.

It is expected that development and refinement of automatic pedestrian detectors will continue inthe United States because of their potential benefits. Improvements are needed in detection accuracy toreduce the number of false alarms and missed calls at intersections. In the meantime, the sensors appearto be best suited for actuated signals at mid-block crossings where there is not a problem with right-turning vehicles producing false calls. It also appears best to have a buffer area separating the sidewalkand the street to reduce the number of false actuations from pedestrians walking along the sidewalk whodo not intend to cross the street. Further research is also needed to test the effectiveness of automateddetection systems without any push button calls.

None of the study sites were characterized by a complete absence of pedestrians crossing on aflashing or steady Don’t Walk signal. This goes to show that neither the use of a push button nor the useof a push button in conjunction with automatic detection can ensure complete control over pedestriancrossing behavior. However, this study showed that the use of automatic detection can result in positivechanges in pedestrian behavior during critical portions of the crossing sequence, and in so doing, willlikely decrease the chance of a pedestrian-motor vehicle conflict.

As noted above, this study evaluated automated detectors at four intersections in three cities. The reader is advised that conditions specific to the study sites and / or the dates that videotaping wasconducted could have affected the results. It is recommended that data be collected at a larger numberof locations in the future as more of these devices are installed in U.S. cities. This will assist trafficengineers in better understanding the types of locations where automated detectors are best suited.

ACKNOWLEDGMENTS

This study was conducted as part of a research project funded by the Federal HighwayAdministration, under contract number DTFH61-92-C-00138. Carol Tan Esse served as theContracting Officer’s Technical Representative. The authors gratefully acknowledge the researchassistance provided by Brian Gallagher (City of Los Angeles), the City of Phoenix Traffic Engineeringstaff, and the Monroe County (Rochester) Department of Transportation. The authors wish to thank thenumerous individuals who collected data. Bradley Keadey, Eric Rodgman, and J. Richard Stewart (allfrom University of North Carolina Highway Safety Research Center) made data analysis possible.

17

APPENDIX

Evaluation of the SMARTWALK 1400 Microwave Pedestrian Sensor in Phoenix, Arizona

The City of Phoenix tested a Smartwalk 1400 Pedestrian Sensor made by Microwave Sensors,Inc., in order to determine if it could be used to supplement or replace the traditional pedestrian pushbutton to call the Walk signal.

Slightly more than half of the 850 traffic signals in Phoenix are equipped with pedestrian pushbuttons. Most of these are at intersections of arterial and collector or local streets outside of thedowntown core. Intersections with pedestrian push buttons all have prominent signing at theintersection stating Push Button for Walk (with an arrow pointing in the direction of travel), and manyhave supplemental signs explaining the meaning of the Walk, flashing Don’t Walk, and steady Don’tWalk signal indications. All intersections have been checked to ensure that the pedestrian push buttonsare conveniently located at the intersection and are accessible. Still, a relatively high proportion ofpedestrians fail to use the push button before crossing.

Field testing was conducted at the intersection of Central and Earll Drive in central Phoenix. Allpedestrians are required to cross Central Avenue on the south leg crosswalk. Pedestrians must use thepush buttons on the southwest or southeast corners of the intersection to obtain a Walk signal andenough time to cross Central Avenue. Without pedestrian actuation, pedestrians will usually not haveenough time to cross Central Avenue. The pedestrian detector was installed at the southwest corner ofCentral Avenue and Earll Drive to detect eastbound pedestrians waiting to cross Central Avenue. Thedetector was set up to supplement the pedestrian push button at the corner.

Instead of hooking up the sensor directly to the traffic signal, it was connected to a red indicatorlight that revealed when it was "activated" by a pedestrian waiting to cross Central Avenue. Dots werelightly painted on the sidewalk to outline the edges of the detection zone. This allowed an observer todetermine if a pedestrian entered the detection zone. Signal engineers decided to use a 4-s delay todetect a pedestrian waiting to cross. It was thought that a 4-s delay was optimal since people walkingacross the detection zone have enough time to clear the zone without actuating the sensor.

The first problem encountered with the microwave sensor was that it would "detect" slow-moving cars making a right turn outside the pedestrian detection zone and cause an activation. This wasa “false call.” To limit the number of false calls, the edge of the detection zone was moved back toabout 76 cm (30 in) behind the curb. Despite moving the detection zone back from the street, someslow-moving, right-turn vehicles (usually Right Turn on Red vehicles) were still creating some falseactivations. Furthermore, with a 76-cm (30-in) setback between the curb and detection zone, aneastbound pedestrian waiting to cross Central Avenue can walk through the detection zone and thenstand at the curb without causing the microwave sensor to be activated.

18

Another hurdle was that some pedestrians would intentionally stand in the shade near the trafficsignal while waiting to cross, and would be outside the pedestrian detection zone. This is especially trueduring the hot summer months when pedestrians are looking for any possible relief from the directsunlight. Because these “shade areas” vary by the time of day, it would be difficult to target any specificarea to include in the detection zone. Because of this problem, it was felt that the microwave sensormay not be a complete replacement for the pedestrian push button, but there is still considerablepotential for its use as a supplement to the push button.

During preliminary observation, yet another limitation was observed: the microwave motionsensor cannot detect which direction a pedestrian is traveling. The goal is to only detect eastboundpedestrians waiting to cross, and not westbound pedestrians who had already crossed, nornorthbound/southbound pedestrians who were walking along the sidewalk through the detection zone. Thus, it is important that the delay be long enough to exclude these other pedestrians walking at thecorner and target the population of eastbound pedestrians. Yet, long delay times may cause the sensorto not detect some pedestrians wanting to cross.

The microwave sensor worked very well if a single pedestrian or small group of pedestrianswere walking north, south, or westbound through the detection zone. They generally would not activatethe sensor. However, if two or more pedestrians occupied any portion of the zone for the 4-s delayinterval in any direction, a call would be sent to the signal, indicating another source of false calls.

The sensor was installed in August 1998. Manual observations of the microwave sensoractivation were conducted in September and October 1998. An observer would discreetly stand nearthe southwest corner in a position where they could observe if the red indicator light activated, and if apedestrian was standing in the detection zone. Eight hours and 25 minutes of manual observations weregathered. A proper detection was recorded if one or more pedestrians waiting to cross Central Avenue(eastbound) stood withing the detection zone for at least 4 s, and the indicator light revealed anactivation. A false call was recorded if the indicator light appeared, which was caused by one or moreother or incorrect types of activations (right-turning vehicles, westbound pedestrians, north/southpedestrians, and other problems). Missed calls were counted when one or more eastbound pedestrianswere waiting to cross (for at least 4 s) but no activations occurred. A summary of the results is shownbelow:

C Proper pedestrian detection and activation = 194C False calls = 182C Missed calls = 27

A proper pedestrian detection may result from a single pedestrian or a group of pedestrianswaiting to cross. More than one activation could be recorded during a given cycle. If a pedestriancrossed eastbound against the signal or during the Walk signal and did not wait at least 4 s beforecrossing, that observation was discounted.

19

Most of the missed calls involved eastbound pedestrians standing between the curb and thedetection zone who had walked through the detection zone in less than 4 s. Thus it is important that thedetection be as close to the edge of the street as possible. A few missed calls involved pedestriansstanding outside the detection zone in other areas. There were no instances of pedestrians standingwithin the detection zone for at least 4 s and not being detected.

The number of false calls was also a concern. On some occasions, it was difficult for theobserver to explain exactly what action may have caused a false call, and a few of the false calls couldnot be explained by any action occurring within or near the detection zone. About 25 of the false callsoccurred because of right turns or the presence of buses/large trucks in the curb lane. When cityofficials contacted the manufacturer’s technical expert, they were asked to send a photograph of theinstallation to see if he could determine a reason why these problems were occurring. He noted that themicrowave sensor was not properly mounted with the respect to the position of the antenna andrecommended rotating the sensor so that the antenna was on the side of the sensor, and not on the topof the microwave sensor.

The rotation of the antenna reduced the number of false calls for right-turning cars and allowedthe placement of the detection zone closer to the intersection. The detection zone was moved to about46 cm (18 in) from the back of the curb, which would also reduce the number of missed calls. Asufficient number of manual observations was not obtained with the revised sensor installation becausethe test period ended and the sensor equipment had to be returned. However, during 30 minutes ofobservation, there were 15 proper detections observed, 13 false calls, and 3 missed calls frompedestrians waiting on the curb. Three of the false calls were caused by right-turning vehicles.

The sensor’s performance was also observed during a steady downpour—the sensor producedcontinuous calls for actuation despite there being no pedestrians present. The manufacturer’s technicalexpert claimed that this should not occur during a light rain, but the manufacturer has not been able toeliminate the problem of continuous calls during a heavy rain.

The manufacturer reported that the sensor installation and antenna are being modified to furtherreduce the number of false calls resulting from right-turning vehicles at the corner. Further testing alsoneeds to be conducted on the optimal delay time before a detection can occur. This optimal delay timemay vary based on the cycle length and the number of pedestrians who are crossing.

At the present time, the sensor appears to be better suited for actuated signals at mid-blockcrossings where there is not a problem of right-turning vehicles that may produce a false call. It is alsobest to have a buffer area (separation) between the sidewalk and the street so that pedestrians walkingalong the sidewalk do not produce false calls.

20

REFERENCES

Beckwith, D.M., and Hunter-Zaworski, K.M. Passive Pedestrian Detection at Unsignalized Crossings. Presented at the 77th Annual Meeting of the Transportation Research Board, Washington, DC,January 1998.

Bentzen, B.L., and Tabor, L.S. Accessible Pedestrian Signals. Report presented to the U.S. AccessBoard, August 1998.

Catchpole, John. Further Development of the Puffin Pedestrian Crossing. VicRoads, Victoria,Australia, November 1996.

Ekman, L., and Draskocsky, M. Trials with Microwave Detection of Vulnerable Road Users andPreliminary Empirical Model Test. University of Leeds, Institute for Transport StudiesWorking Paper No 336, Leeds England, 1992.

King, Wayne. Pelicans, Puffins, and Kiwis. Conference Proceedings IPENZ Annual Conference,Volume 2. Institute of Professional Engineers, New Zealand, 1994.

Levelt, P.B.M. Improvement of Pedestrian Safety and Comfort at Traffic Lights: Results fromFrench, British, and Dutch Field Tests. Institute for Road Safety Research, Leidschendam,Netherlands, 1992.

Manual on Uniform Traffic Control Devices for Streets and Highways. Federal HighwayAdministration, Washington, DC, 1988.

Sherborne, D.J. Existing Techniques for Detecting Vulnerable Road Users. Drive II ProjectV2005 VRU-TOO, 1992.

Tan, C., and Zegeer, C.V. European Practices and Innovations for Pedestrian Crossings. ITE Journal,Vol. 65, No. 11, pp. 24-31, November 1995.

UK Department of Transport. Network Management Advisory Leaflet: The Use of PuffinPedestrian Crossings. Network Management and Driver Information Division, 1993.

Zegeer, C.V., Cynecki, M., Fegan, J., Gilleran, B., Lagerwey, P., Tan, C., Works, B., andTransportation Technology Evaluation Center. FHWA Study Tour for Pedestrian andBicyclist Safety in England, Germany, and The Netherlands. Report FHWA-PL-95-006. Federal Highway Administration, Washington, DC, 1994.

21

Zegeer, C., Opiela, K., and Cynecki, M. Pedestrian Signalization Alternatives (Final Report),Federal Highway Administration, Washington, DC, July 1985.